AI-Driven Digital Twins for Tasks Offloading in 6G UAV-Aided MEC Networks.

New generation networks are expected to enable a novel plethora of disruptive delay-sensitive applications mainly supported by edge computing capabilities in different domains. In such a novel scenario the use of the digital twin (DT) technology appears to offer meaningful opportunities to tailor the network to the service needs. This paper aims at proposing the use of the DT paradigm in a Unmanned Aerial Vehicle (UAV) aided ground Multiaccess Edge Computing (MEC) network, where a set of edge node DTs monitor and evaluate the states of physical edge nodes and a DT of the entire edge computing network provides training data to be exploited by machine learning to perform decisions about offloading strategies, and assignment of a shared UAV with EC on board capabilities to congested areas, aiming at minimizing the worst completion tasks delay. In this context, a matching game has been developed to perform these decisions, while the network congestion has been analyzed and learned by applying an echo-state network fed from the DT of the system. Finally, extensive performance analysis through system simulations has been conducted, confirming the validity of the framework proposed, as long as the benefits deriving by its adoption in terms of completion task delay, prediction accuracy, mean channel interference and network congestion reduction.